Language generation is often used by artificial intelligences, such as chat bots, to conduct a conversation with a human user. A goal of language generation is to make the language sound as natural as possible. Unfortunately, existing techniques for language generation often produces language that sounds artificial or “canned.”
This is especially true when the artificial intelligence is asked to express an opinion on a topic (e.g., whether a given movie is good or not). Conventionally, opinion expression is handled using templates, in which predefined responses are provided. The predefined responses may include variables into which details of the response can be inserted. Because these intelligences use predefined responses, each response tends to sound the same and the intelligence is typically viewed as robotic, lacking a personality of its own.